Right before the start of the Iran war, I asked DeepSeek what the odds were, in the event the Trump administration decided to go to war with Iran, that the Trump administration would succeed in replacing the then current Iranian regime with one the Trump administration would find tolerable. DeepSeek said the odds were basically zero. Maybe the Trump administration wasn't using AI enough.
If one asks intelligent questions, one has some hope of getting an intelligent answer.
In this case, I think the questions were more like give me a list of 10 things we can bring down.
And the girl's school was in there in the list even though the compound had ceased to be a IRGC one for about 14 years now and the roof of the school had a visible (from air) blue marker on it to indicate it is a school.
Do you think palantir was involved in the targeting AI? Because the data on the girls school was 10 years out of date. Apparently it used to be some kind of naval administration office. AI in general can't seem to grasp the distinctions imbedded in recency and superiority. Humans are constantly weighing these iterative contexts. And you know --government, education and companies have no immunity to these highly-funded, sophisticated and overpromising sales pitches by huge salesforces, advertising and PR hype. Wish someone would send you a pitch deck!
My understanding is that Palantir does not handle the algorithmic/logic/decision making end of these things. They are just pipes that are now everywhere. AI is the brains. Palantir does not provide AI. In this case, the AI was Claude. Even though Claude is supposed to be out, it is still integrated into the wrapper that Palantir uses, I hear. Palantir indicated that they can just as well wrap ChatGPT.
The problem, as I see it, is that the prompt is legible in every output. If you ask an LLM something for which there's no sharply delineated and obvious answer (and frankly, sometimes even then) you'll get one that assumes the validity of the premise.
You might be able to steer this a little with some thoughtful pre-prompting, but that is not reliable. The models routinely "ignore" pre-prompts.
When you use an LLM for actual work, you have to be vigilant in managing the context window. Therein lies one of the central weaknesses of this technology: the inability to "see" the forest for all the trees.
And the closer you get to the edge of the training distribution, which will be true for detailed questions about anything that hasn't happened before, the answers get increasingly, ahem, inventive. And not in a way that has any relationship to reality.
For a military application, this makes it worse than useless.
Whether or not this is is an accurate depiction of what happened, these questions need to be answered, especially “What testing did you do before you rolled LLMs into the targeting process?”
A lot of the public discussion imagines a general asking a chatbot what to bomb. That is probably the wrong picture. The more plausible danger is AI being used to process intelligence, prioritise targets, and accelerate the kill chain in ways that give an illusion of precision while hiding bad data, weak assumptions, and reduced human scrutiny. The hysteria is misplaced in one sense, but the underlying concern is not.
“ This is the recurring illusion of overequipped leaders: Because they can map the battle space, they think they understand the war. But war is never merely a technical contest. It is shaped by grievance, sacred narrative, the memory of past humiliations and the desire for revenge. Those are not atmospheric complications added to an otherwise technical enterprise. They are what the war is about.”
I doubt very much that AI was used at the strategy level. Hegseth and Trump already knew what they wanted to do which was simply walk-in, use violence, and get what they want.
Those two idiots are incapable of strategy and they mos def are not getting what they wanted, which, if you recall, was originally TOTAL UNCONDITIONAL SURRENDER. The only people who have now unconditionally surrendered are the two very low IQ individuals who failed to follow Harry Truman's advice to "Never kick a wet cow paddy on a hot day" and are now expecting NATO members to clean up the disastrous mess they left.
They could both have learned a lot from Genghis Khan, who subjugated more territory in 25 years than the Romans did in 400: [from the book "Genghis Khan and the Making of the Modern World," by Jack Weatherford]
The first key to leadership is self control, particularly the mastery of pride. He warned them "If you can't swallow your pride, you can't lead." He admonished them never to think of themselves as the strongest or smartest. He warned his sons not to talk too much. "[a ruler] can never be happy until his people are happy." He warned them against the pursuit of a "colorful" life with material frivolities and wasteful pleasures. In that case, "you will be no better than a slave, and you will surely lose everything ... You may conquer an army with superior tactics and men, but you can conquer a nation only by conquering the hearts of the people."
About your last paragraph: "Win the war, lose the Peace". Even if you go "full in", having your soldiers patrolling every town and city of your enemy, will not help, if you cannot convince the conquered that their chances are better with you, than with the previous rulers.
I am reading "The Book of Why" by Pearl and Mackenzie. Using Pearl's paradigm, strategy is a level 3 issue, well above mere data analysis, and this is a situation where AI yet has no role, because it involves counterfactuals. By the way, kudos for Mr. Marcus, mentioned in the book.
"There is a whole bunch of people (Mostly non-tech) having posts for decision making that think AI is intelligence and they would trust the tech, including when it says you are absolutely right."
It double validates it. First, they trust AI, and then AI tells them how right they are.
Why did Trump fill up his most key positions with actors? Pete Hegseth is an embarrassing joke. His tattoos are disgustingly a form of idol worship. His first request was for his very own TV studio to rain propaganda on us twenty four seven. Kristi Noem has the same problem. We watched Biden do it.
Trump is the equivalent of a Hollywood style actor. We’re watching a movie set full of a bunch of useful idiots.
It’s a clown show meant to keep us entertained while the destruction of our country continues. It’s absolutely unbelievable to watch.
“Why did Trump fill up his most key positions with actors?”
Trump’s single success (remember he drove 6 businesses into bankruptcy, one of which was a casino FFS!) was in a reality TV show, and ever since he’s treated the world around him, elections, administrations, and wars, as reality tv, with him as producer and star.
AI lacks discernment to choose the right experts to elevate. Yashar Ali is an Iranian American journalist with excellent sources. He had said for years that the Islamic Republic could not be easily toppled, in spite of his wishes and the wishes of other Iranians both in and outside the country. Listening to one person who knows what they are talking about will beat a sycophantic chatbot pretty frequently.
What they had was a desire to control energy worldwide (see: Venezuela, drone attacks on Russian oil facilities, seizure of "shadow oil tanker fleets", etc.), with emphasis on denying oil to China (most Venezuelan, Russian and Iranian oil goes to China) to cripple China's economy.
And as Trump's boggle-headed line went, "When the oil price goes up, we make a lot of money." He gave away the game, not that it took much analysis given the Iraq war twenty-five years ago which was undertaken because Saddam Hussein was monkeying with the OPEC oil price.
None of that was going to work. Everyone outside the administration with any knowledge of the economic and military balance of power and the capabilities of Iran knew it wouldn't work. Unfortunately the moron US electorate elected someone even more moronic than they were, and he put into his cabinet people who were equally moronic.
Now we're headed into a global recession and possibly a global depression.
Over the past week or two, Professor Robert Pape from the University of Chicago, who has studied escalation in war for the last thirty years, has been on just about every TV and YouTube show out there explaining what he calls "the escalation trap." He says the odds are 75% that Trump goes for a ground war. That, he says, will be Stage 3 on his five-stage theory and a major blunder.
Follow the $$, good results = "done by AI", bad results = "done by people". There is no way to know as all this is and will be secret. But we can follow the patterns of the "news" we are allowed to see.
When there are a lot of money in balance people, some of them, will take the fall.
No wonder Hegseth is banning reporters and journalists from talking to people in the Pentagon. He cannot have the media exposing what a clusterf**k this war is turning out to be.
Romero Lupascu: . . . Follow the $$, indeed. And this doesn't even take into account the universal gambling that is going on, and their potential threads with what Trump et al happen to do, regardless of other more normative considerations (like world order or our troops). Who cares what else happens when you can move the chessmen around and bet on those moves ahead of time?
AI is then perfect for this administration. I hear people are in dread of getting fired for being bearer of bad news. So, they tell POTUS what he wants to hear. That might explain the surprises about resiliency of Iran.
Only unanswered question is then who is more sycophantic? AI or officials?
operations vs strategy distinction is interesting but... isn't that just what we'd expect? like who actually thought a general was going to let an AI make the call lol
Right before the start of the Iran war, I asked DeepSeek what the odds were, in the event the Trump administration decided to go to war with Iran, that the Trump administration would succeed in replacing the then current Iranian regime with one the Trump administration would find tolerable. DeepSeek said the odds were basically zero. Maybe the Trump administration wasn't using AI enough.
GIGO.
If one asks intelligent questions, one has some hope of getting an intelligent answer.
In this case, I think the questions were more like give me a list of 10 things we can bring down.
And the girl's school was in there in the list even though the compound had ceased to be a IRGC one for about 14 years now and the roof of the school had a visible (from air) blue marker on it to indicate it is a school.
Do you think palantir was involved in the targeting AI? Because the data on the girls school was 10 years out of date. Apparently it used to be some kind of naval administration office. AI in general can't seem to grasp the distinctions imbedded in recency and superiority. Humans are constantly weighing these iterative contexts. And you know --government, education and companies have no immunity to these highly-funded, sophisticated and overpromising sales pitches by huge salesforces, advertising and PR hype. Wish someone would send you a pitch deck!
My understanding is that Palantir does not handle the algorithmic/logic/decision making end of these things. They are just pipes that are now everywhere. AI is the brains. Palantir does not provide AI. In this case, the AI was Claude. Even though Claude is supposed to be out, it is still integrated into the wrapper that Palantir uses, I hear. Palantir indicated that they can just as well wrap ChatGPT.
The problem, as I see it, is that the prompt is legible in every output. If you ask an LLM something for which there's no sharply delineated and obvious answer (and frankly, sometimes even then) you'll get one that assumes the validity of the premise.
You might be able to steer this a little with some thoughtful pre-prompting, but that is not reliable. The models routinely "ignore" pre-prompts.
When you use an LLM for actual work, you have to be vigilant in managing the context window. Therein lies one of the central weaknesses of this technology: the inability to "see" the forest for all the trees.
And the closer you get to the edge of the training distribution, which will be true for detailed questions about anything that hasn't happened before, the answers get increasingly, ahem, inventive. And not in a way that has any relationship to reality.
For a military application, this makes it worse than useless.
Whether or not this is is an accurate depiction of what happened, these questions need to be answered, especially “What testing did you do before you rolled LLMs into the targeting process?”
https://houseofsaud.com/iran-war-ai-psychosis-sycophancy-rlhf/
Holy Cow! You are right.
A lot of the public discussion imagines a general asking a chatbot what to bomb. That is probably the wrong picture. The more plausible danger is AI being used to process intelligence, prioritise targets, and accelerate the kill chain in ways that give an illusion of precision while hiding bad data, weak assumptions, and reduced human scrutiny. The hysteria is misplaced in one sense, but the underlying concern is not.
There was an article about this very subject in the New York Times yesterday: https://www.nytimes.com/2026/03/29/opinion/israel-us-war-iran-literature.html?unlocked_article_code=1.XlA.WkKB.mEbi3eFln6NH&smid=url-share
Apparently this is exactly what the Israelis did.
“ This is the recurring illusion of overequipped leaders: Because they can map the battle space, they think they understand the war. But war is never merely a technical contest. It is shaped by grievance, sacred narrative, the memory of past humiliations and the desire for revenge. Those are not atmospheric complications added to an otherwise technical enterprise. They are what the war is about.”
What a good article .
I doubt very much that AI was used at the strategy level. Hegseth and Trump already knew what they wanted to do which was simply walk-in, use violence, and get what they want.
Those two idiots are incapable of strategy and they mos def are not getting what they wanted, which, if you recall, was originally TOTAL UNCONDITIONAL SURRENDER. The only people who have now unconditionally surrendered are the two very low IQ individuals who failed to follow Harry Truman's advice to "Never kick a wet cow paddy on a hot day" and are now expecting NATO members to clean up the disastrous mess they left.
They could both have learned a lot from Genghis Khan, who subjugated more territory in 25 years than the Romans did in 400: [from the book "Genghis Khan and the Making of the Modern World," by Jack Weatherford]
The first key to leadership is self control, particularly the mastery of pride. He warned them "If you can't swallow your pride, you can't lead." He admonished them never to think of themselves as the strongest or smartest. He warned his sons not to talk too much. "[a ruler] can never be happy until his people are happy." He warned them against the pursuit of a "colorful" life with material frivolities and wasteful pleasures. In that case, "you will be no better than a slave, and you will surely lose everything ... You may conquer an army with superior tactics and men, but you can conquer a nation only by conquering the hearts of the people."
About your last paragraph: "Win the war, lose the Peace". Even if you go "full in", having your soldiers patrolling every town and city of your enemy, will not help, if you cannot convince the conquered that their chances are better with you, than with the previous rulers.
Singing "Daisy" in the Sit Room.
And seeing pie all over Trump’s and Hegseth’s face.
I am reading "The Book of Why" by Pearl and Mackenzie. Using Pearl's paradigm, strategy is a level 3 issue, well above mere data analysis, and this is a situation where AI yet has no role, because it involves counterfactuals. By the way, kudos for Mr. Marcus, mentioned in the book.
"There is a whole bunch of people (Mostly non-tech) having posts for decision making that think AI is intelligence and they would trust the tech, including when it says you are absolutely right."
It double validates it. First, they trust AI, and then AI tells them how right they are.
This is just the fall of morality in tech.
they build what never should have been built.
Why did Trump fill up his most key positions with actors? Pete Hegseth is an embarrassing joke. His tattoos are disgustingly a form of idol worship. His first request was for his very own TV studio to rain propaganda on us twenty four seven. Kristi Noem has the same problem. We watched Biden do it.
Trump is the equivalent of a Hollywood style actor. We’re watching a movie set full of a bunch of useful idiots.
It’s a clown show meant to keep us entertained while the destruction of our country continues. It’s absolutely unbelievable to watch.
God help us all against these evil evil people.
In a Pathocracy those in charge can't help but turn to others equally as psychotic. All others are outside their psychological event horizon.
“Why did Trump fill up his most key positions with actors?”
Trump’s single success (remember he drove 6 businesses into bankruptcy, one of which was a casino FFS!) was in a reality TV show, and ever since he’s treated the world around him, elections, administrations, and wars, as reality tv, with him as producer and star.
AI lacks discernment to choose the right experts to elevate. Yashar Ali is an Iranian American journalist with excellent sources. He had said for years that the Islamic Republic could not be easily toppled, in spite of his wishes and the wishes of other Iranians both in and outside the country. Listening to one person who knows what they are talking about will beat a sycophantic chatbot pretty frequently.
In the MAGA mind, expertise is considered suspect.
I don't think the humans had a strategy.
What they had was a desire to control energy worldwide (see: Venezuela, drone attacks on Russian oil facilities, seizure of "shadow oil tanker fleets", etc.), with emphasis on denying oil to China (most Venezuelan, Russian and Iranian oil goes to China) to cripple China's economy.
And as Trump's boggle-headed line went, "When the oil price goes up, we make a lot of money." He gave away the game, not that it took much analysis given the Iraq war twenty-five years ago which was undertaken because Saddam Hussein was monkeying with the OPEC oil price.
None of that was going to work. Everyone outside the administration with any knowledge of the economic and military balance of power and the capabilities of Iran knew it wouldn't work. Unfortunately the moron US electorate elected someone even more moronic than they were, and he put into his cabinet people who were equally moronic.
Now we're headed into a global recession and possibly a global depression.
Over the past week or two, Professor Robert Pape from the University of Chicago, who has studied escalation in war for the last thirty years, has been on just about every TV and YouTube show out there explaining what he calls "the escalation trap." He says the odds are 75% that Trump goes for a ground war. That, he says, will be Stage 3 on his five-stage theory and a major blunder.
Here's a video with him today:
https://www.youtube.com/watch?v=wP2Za1sqhek
You're right that the humans had not strategy, esp. those at the top -- they still don't!
So they reverted to some combination of LLM's and Palantir and who knows what else
*no strategy*
Follow the $$, good results = "done by AI", bad results = "done by people". There is no way to know as all this is and will be secret. But we can follow the patterns of the "news" we are allowed to see.
When there are a lot of money in balance people, some of them, will take the fall.
Nothing new.
Just MHO.
No wonder Hegseth is banning reporters and journalists from talking to people in the Pentagon. He cannot have the media exposing what a clusterf**k this war is turning out to be.
Romero Lupascu: . . . Follow the $$, indeed. And this doesn't even take into account the universal gambling that is going on, and their potential threads with what Trump et al happen to do, regardless of other more normative considerations (like world order or our troops). Who cares what else happens when you can move the chessmen around and bet on those moves ahead of time?
AI is then perfect for this administration. I hear people are in dread of getting fired for being bearer of bad news. So, they tell POTUS what he wants to hear. That might explain the surprises about resiliency of Iran.
Only unanswered question is then who is more sycophantic? AI or officials?
Sounds remarkably like the situation in Hitler's bunker in 1945, but with added Artificial Stupidity.
I don’t think any of them has the intelligence or imagination to ask, “Are we the bad guys,” let alone answer the question honestly.
Wow, I was thinking about the same scenes of "Der Untergang" myself :-)
"Mit dem Angriff Steiner's wird dies alles in Ordnung kommen".
No support from army group "Steiner" for Adolf, no miraculous help for idiots Pete and DT.
operations vs strategy distinction is interesting but... isn't that just what we'd expect? like who actually thought a general was going to let an AI make the call lol